ports/openmpi
ports/openmpi
1
1
Форкнуть 0
Код Релизы Активность
3 455 коммитов 3 Ветки 0 Теги 130 MiB
d4434bac91
Граф коммитов

3455 Коммитов

This Branch
This Branch
Все ветки
Автор SHA1 Сообщение Дата
Jeff Squyres
8d4506a03e Move win_examples out into tmp tree 
This commit was SVN r3649.
 2004-11-21 22:32:22 +00:00
Brian Barrett
fa6558db61 * update RMS pcm to match changes to gpr / ns interfaces 
This commit was SVN r3648.
 2004-11-21 17:23:58 +00:00
Ralph Castain
1ccb5cde4d Cleanup some of the warnings reported by George. 
This commit was SVN r3647.
 2004-11-21 17:20:42 +00:00
Ralph Castain
73844230bc Fix an initialization problem that valgrind complained about - forgot to mirror the fix that was done for the replica. 
This commit was SVN r3646.
 2004-11-21 04:41:51 +00:00
Ralph Castain
48b0b10464 For some unknowable reason, svn refused to pickup the ns subtree. Checked the whole thing out again, copied the correct files on top of the old ones, and now recommitting them. 
This commit was SVN r3645.
 2004-11-20 19:20:13 +00:00
Ralph Castain
bf9087d9d1 The merged main trunk and gpr integration branch. Tested on Mac only so far - will check out and test on Linux. If that has a problem, will back all changes out (again), but I think we have this one correct. Will send out a more complete change notice once testing is complete. 
This commit was SVN r3644.
 2004-11-20 19:12:43 +00:00
George Bosilca
c59d37315e Remove all debugging stuff (all those macro not C99 compliant). 
And remove all compilation warnings.

This commit was SVN r3643.
 2004-11-20 00:18:38 +00:00
George Bosilca
29652b487a Dont forget to correctly initialize the PTL in the not threaded case. Now GM work in all cases. One step forward ... 
This commit was SVN r3642.
 2004-11-19 22:33:30 +00:00
George Bosilca
02820f0d3b Separate the send request and the DMA memory in 2 lists. Now they are independent as we dont always need a send request to be able to send a message to a peer. 
This commit was SVN r3641.
 2004-11-19 22:27:03 +00:00
George Bosilca
a777f4ad18 Use only types with specified length (int32_t, uint32_t ...) 
This commit was SVN r3640.
 2004-11-19 21:35:15 +00:00
George Bosilca
452920ed6c Correctly handle the free list grow. 
This commit was SVN r3639.
 2004-11-19 21:34:17 +00:00
George Bosilca
1dfcf38d3c Atomically increase the size of the req_bytes_sent. 
Decrease and limit the locked part of the function.

This commit was SVN r3638.
 2004-11-19 21:33:25 +00:00
Craig E Rasmussen
5c17dfda0e Directory to hold shell scripts to generate compiler dependent Fortran code. 
This commit was SVN r3633.
 2004-11-18 23:58:42 +00:00
Craig E Rasmussen
315e17f6ea Removed MPI_Wtick and MPI_Wtime as they are declared in mpif.h 
This commit was SVN r3632.
 2004-11-18 23:56:05 +00:00
Craig E Rasmussen
0eaa017e36 Finished with one complete pass through the xml. Now need to 
make sure it leads to correct code generation.

This commit was SVN r3631.
 2004-11-18 23:01:32 +00:00
George Bosilca
69efbe09ea Decrease the latency. Report the send request as completed as soon as the data is in the GM buffers. 
This commit was SVN r3629.
 2004-11-18 20:25:01 +00:00
George Bosilca
7a9f100467 When no more send fragments are available wait until one of them became available. There are 2 ways to solve this problem, and for the alpha release I take the easiest one. Read the comment in the file for more informations. 
This commit was SVN r3628.
 2004-11-18 20:24:16 +00:00
George Bosilca
bcd498a2e1 Use the correct function to initialize the fragment header. 
This commit was SVN r3627.
 2004-11-18 20:22:29 +00:00
George Bosilca
a186a02480 Limit the number of items in the GM free lists 
This commit was SVN r3626.
 2004-11-18 20:21:25 +00:00
George Bosilca
394e5abd04 Create function to specifically initialize the header depending on the communication type. 
This commit was SVN r3625.
 2004-11-18 20:20:42 +00:00
Brian Barrett
af1f8e485a * interupt polling for an incoming signal right away by adding an event for 
a pipe.  This eliminates the 2-3 second delay in signal delivery that
  was making using libevent for cntl-c in mpirun a bit evil.

This commit was SVN r3620.
 2004-11-18 16:09:59 +00:00
George Bosilca
67d6d86638 Back to working state using the new headers. Having less data to move around decrease 
a little bit the latency. But I still have to improuve it ...

This commit was SVN r3618.
 2004-11-18 06:36:26 +00:00
George Bosilca
b80e69bc69 It's a free list so let's use it as a free list ! 
This commit was SVN r3617.
 2004-11-18 06:34:28 +00:00
Rich Graham
72f9b1df14 remove the shared memory control backing file on 
termination.  remove some debug code.

This commit was SVN r3615.
 2004-11-18 04:18:33 +00:00
Brian Barrett
5ee42c2a58 * correct the installation paths of a bunch of headers 
* make sure to install the libevent header
* make Automake create the sysconf_DATA directory before trying to
  install the config files

This commit was SVN r3614.
 2004-11-18 03:53:42 +00:00
Tim Woodall
be253609ac changes to support running a non-threaded build on oversubscribed processors 
This commit was SVN r3613.
 2004-11-18 01:58:30 +00:00
Brian Barrett
fde966be5f * install the default hostfile and mca params files only if there isn't 
already one there.  Credit to Jeff from LAM for the code

This commit was SVN r3611.
 2004-11-18 00:55:03 +00:00
George Bosilca
89b6b3b66a Keep the correct pointer to the DMA memory. 
This commit was SVN r3610.
 2004-11-18 00:26:20 +00:00
George Bosilca
33426c5af1 Just a checkpoint. Now compile with the new header. 
Improuvement:
- latency went down by 15 us. It's still a little bit hist 12 but I'm working on that.
- correctly initialize and cleaups independing on the number of myrinet cards.
- associate one PTL by card and one thread by PTL if we are in a MT environment.

This commit was SVN r3609.
 2004-11-18 00:23:19 +00:00
Tim Woodall
89978d698b correct ntoh/hton macro 
This commit was SVN r3607.
 2004-11-17 23:43:30 +00:00
Tim Woodall
c80818463f use oob barrier for mpi init 
This commit was SVN r3606.
 2004-11-17 23:37:49 +00:00
Tim Woodall
f78fb7a029 oops - should default to off for non-threaded build 
This commit was SVN r3605.
 2004-11-17 23:21:18 +00:00
George Bosilca
da76409962 When we copy a convertor just copy the minimum required fields. All others will be 
initialized in the ompi_convertor_init_for_* functions.

This commit was SVN r3604.
 2004-11-17 22:59:48 +00:00
Tim Woodall
9cb5f05ec1 comments 
This commit was SVN r3603.
 2004-11-17 22:54:43 +00:00
Tim Woodall
70d1c1aafd rather large commit: 
- change mca_ptl_base_header_t definition to decrease the
  header size for small messages. note that this requires
  all ptls to be updated. tcp/self/sm/mx have been changed,
  gm/ib/quadrics will be broken by this commit. george and
  mitch have volunteered to make the required changes to gm/ib
- revised matching logic to reduce function call overhead
- changes to tcp/self/sm/mx ptls to support the revised headers

This commit was SVN r3602.
 2004-11-17 22:47:08 +00:00
Tim Woodall
4ac1cc5c3a allow progress flag to be turned off only if it hasn't been reset 
This commit was SVN r3601.
 2004-11-17 21:00:27 +00:00
Rich Graham
7755ed6ad4 inline a few more functions. 
This commit was SVN r3600.
 2004-11-17 19:25:22 +00:00
Brian Barrett
24f2e57da2 * add changes subversion for some reason didn't schedule to be re-added 
when rolling back r3589

This commit was SVN r3598.

The following SVN revision numbers were found above:
  r3589 --> open-mpi/ompi@9c85e6be5f
 2004-11-17 12:37:27 +00:00
Jeff Squyres
582ac8f3ba Added generated script to svn:ignore 
This commit was SVN r3597.
 2004-11-17 12:20:21 +00:00
Jeff Squyres
e423d2107a Minor update. 
This commit was SVN r3596.
 2004-11-17 11:31:22 +00:00
George Bosilca
8befccfc46 Remove a compile warning about the array_of_type pointer mismatch. 
This commit was SVN r3595.
 2004-11-17 05:03:05 +00:00
George Bosilca
bdc3c7995a Use the correct type for some of the variables. 
This commit was SVN r3591.
 2004-11-17 03:47:41 +00:00
George Bosilca
9690904dea MPI standard state that MPI_Win_f2c should return MPI_WIN_NULL if no windows correspond to the Fortran value. 
As we dont have windows right now, I still raise the error but I return MPI_WIN_NULL.

This commit was SVN r3590.
 2004-11-17 03:46:56 +00:00
Jeff Squyres
9c85e6be5f New script to submit test results. Not yet complete, but this client 
side seems to be working.  Now to hook up something on the server
side... 

This commit was SVN r3589.
 2004-11-17 02:37:20 +00:00
Brian Barrett
23a6d5bb60 * roll back r3584 (gpr changes to reduce floods) as it appears to cause 
some instability on Linux

This commit was SVN r3587.

The following SVN revision numbers were found above:
  r3584 --> open-mpi/ompi@52add381d0
 2004-11-17 02:30:07 +00:00
George Bosilca
6c9f27a2f5 Mainly cleanups + some variable rename. 
Help the compiler a little bit : add const to all arguments non modified inside the
functions.

This commit was SVN r3586.
 2004-11-16 23:50:31 +00:00
Jeff Squyres
b73c7ae6c9 Add missing <netinet/in.h> 
This commit was SVN r3585.
 2004-11-16 23:12:18 +00:00
Brian Barrett
52add381d0 * Merge over the gpr changes Ralph has made on the gpr-integration branch. 
This may trigger a complete rebuild :(.  Short overview of changes:

  - reduce number of network slams at startup
  - prevent gpr from hanging when doing process death code
  - general gpr cleanups

This commit was SVN r3584.
 2004-11-16 22:53:33 +00:00
Craig E Rasmussen
c21f91582d Initial version. This file template contains type information that 
depends on the Fortran compiler.

This commit was SVN r3583.
 2004-11-16 22:25:18 +00:00
George Bosilca
45e39e033b Correct several bugs in the case when we are in a homogeneous environnement. Now the ddt engine handle 
the contiguous datatype correctly and fast. However in the "developer mode" (OMPI_ENABLE_DEBUG set to 1)
there are additional test that are made. If you plan to look for performances please compile OpenMPI
in normal mode.
Changes:
- correctly generate the stack when we have an initial position
- same thing when we start from ZERO.
- in contiguous mode handle the partial copy of a data (using the stack)
- in the case when not all the data is packed/unpacked leave the stack in a correct state
- take care of data with size ZERO (who really want to create such things !!!)
- general pack/unpack functions handle the new stack representation
- make sure that we never goes outside the user provided buffer

This commit was SVN r3581.
 2004-11-16 21:27:57 +00:00